Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
A weighted routing matrix that assigns AI model tasks to the cheapest model that meets quality thresholds. Scores models across real evaluations, benchmarks, sentiment, and cost, then applies policy rules to pick an effective winner per category. Includes a daily scorecard template to track and update routing decisions.
Combining weighted scoring with explicit swap-only-on-material-delta rules prevents churn and cost creep that comes from routing on benchmarks alone.
When to use it
- Routing complex coding tasks to the highest-scoring code model
- Keeping routine cron jobs on cheaper models to reduce API spend
- Swapping model assignments when a weekly scorecard shows a material score shift
- Enforcing a policy fallback when a preferred provider is temporarily excluded
- Comparing Grok, Gemini, and GPT outputs for sentiment-heavy research tasks
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: which model should handle complex coding tasks this week?
- 1Pull current scorecard scores for the Coding (complex) category
- 2Apply weights: 45% real eval, 30% benchmark, 20% sentiment, 5% cost
- 3Compute raw score for each candidate model
- 4Check policy constraints (e.g., Anthropic exclusion triggers auto-promote of #2)
- 5Return effective #1 with confidence rating
GPT-5.3 Codex assigned as effective model for complex coding with confidence score